Contrary to some of the other reviews on here, my favourite part about Carbon is probably their pricing. They charge based on the content processed than the number of connectors, which was beneficial for us since we had many customers but each customer doesn't have too high of a usage. The most important part for our company was having a data connector that was affordable, and it was a lot cheaper to have Carbon handle all of that than having to set up that system ourselves. I think that might also be their strategy, which is to create value in providing a more affordable option in that piece of using external data to train LLMs. We’ve saved many multiples of what we spent according to our estimates.

Probably their universal retrieval engine. At Rabbu, we need to include unstructured data from many different sources/programs, and the ability to accomplish that through Carbon has allowed us to improve our operational efficiency. Carbon also has SOC 2 Type II compliance, which is also something that we value since our customer confidence depends on our reputation and we try to ensure that all products we use are well-protected.

So one thing to understand is that the core of our AI model is long-term memory, which is comprised of conversations, documents, websites, and any external knowledge over time. The problem arises only when it comes to keeping the uploads of existing data and syncing of external knowledge robust and up-to-date. We were evaluating building in-house solutions based on Langchain, but it was a lot harder than we initially thought. There were bursts of workloads during ad-hoc times, upload speed was not reliable, and the setting up of an entire system was very time-consuming. What we like most about Carbon is kind of like this one-stop-shop for all of our data-connection concerns, at a reliable speed and a reasonable cost.